Contributors: IBM Research Zurich; Laboratoire Nanotechnologies et Nanosystèmes Sherbrooke (LN2); Université de Sherbrooke = University of Sherbrooke (UdeS)-École Centrale de Lyon (ECL); Université de Lyon-Université de Lyon-École Supérieure de Chimie Physique Électronique de Lyon (CPE)-Institut National des Sciences Appliquées de Lyon (INSA Lyon); Université de Lyon-Institut National des Sciences Appliquées (INSA)-Institut National des Sciences Appliquées (INSA)-Centre National de la Recherche Scientifique (CNRS)-Université Grenoble Alpes (UGA); Institut Interdisciplinaire d'Innovation Technologique Sherbrooke (3IT); Université de Sherbrooke = University of Sherbrooke (UdeS); Nanostructures, nanoComponents & Molecules - IEMN (NCM - IEMN); Institut d’Électronique, de Microélectronique et de Nanotechnologie - UMR 8520 (IEMN); Centrale Lille-Université de Lille-Centre National de la Recherche Scientifique (CNRS)-Université Polytechnique Hauts-de-France (UPHF)-JUNIA (JUNIA); Université catholique de Lille (UCL)-Université catholique de Lille (UCL)-Centrale Lille-Université de Lille-Centre National de la Recherche Scientifique (CNRS)-Université Polytechnique Hauts-de-France (UPHF)-JUNIA (JUNIA); Université catholique de Lille (UCL)-Université catholique de Lille (UCL); The authors acknowledge the Binnig and Rohrer Nanotechnology Center(BRNC) at IBM Research Europe - Zurich. Special thanks go to Jean-MichelPortal, Eloi Muhr and Dominique Drouin for their contributions to the de-sign of the NMOS transistors used in this work. The authors also extendtheir gratitude to Stephan Menzel for the fruitful discussions and to RalphHeller for his assistance in wire-bonding the chip. This work was funded by SNSF ALMOND (grant ID: 198612), by the European Union and Swiss state secretariat SERI within the H2020 MeM-Scales (grant ID: 871371), MANIC (grant ID: 861153), PHASTRAC (grant ID: 101092096) and CHIST-ERA UNICO (20CH21-186952) projects; ANR-05-GANI-0005,Mugene,Approche intégrée combinant la génétique, la génomique et la biologie musculaire pour gérer la qualité de la viande bovine selon le potentiel de croissance des animaux et les facteurs d'élevage.(2005); European Project: 871371,H2020-ICT-2018-20,H2020-ICT-2019-2,MeM-Scales(2020)
نبذة مختصرة : International audience ; Analog in-memory computing is an emerging paradigm designed to efficiently accelerate deep neural network workloads. Recent advancements have focused on either inference or training acceleration. However, a unified analog in-memory technology platform-capable of on-chip training, weight retention, and long-term inference acceleration-has yet to be reported. This work presents an all-in-one analog AI accelerator, combining these capabilities to enable energy-efficient, continuously adaptable AI systems. The platform leverages an array of analog filamentary conductive-metal-oxide (CMO)/HfO x resistive switching memory cells (ReRAM) integrated into the back-end-of-line (BEOL). The array demonstrates reliable resistive switching with voltage amplitudes below 1.5 V, compatible with advanced technology nodes. The array's multi-bit capability (over 32 stable states) and low programming noise (down to 10 nS) enable a nearly ideal weight transfer process, more than an order of magnitude better than other memristive technologies. Inference performance is validated through matrix-vector multiplication simulations on a 64 × 64 array, achieving a root-mean-square error improvement by a factor of 20 at 1 s and 3 at 10 years after programming, compared to state-of-the-art. Training accuracy closely matching the software equivalent is achieved across different datasets. The CMO/HfO x ReRAM technology lays the foundation for efficient analog systems accelerating both inference and training in deep neural networks.
Processing Request
No Comments.